Research On RSSI Three-dimensional Fingerprint Localization Algorithm Based On Deep Learning

In current years,indoor positioning has attracted giant attention from academia and industry.Due to the problem of indoor GPS signal strength,indoor positioning machine is usually located near the regarded vicinity factor.While the potential demand for indoor navigation services has been addressed at the industry level,there are still such as the lack of corresponding data quality assessment tools maintenance problems such as limited position precision,especially when dealing with different data sources,the data used in data processing has potential subprime and approximation,so different data sources with different method of reliability and limitation.Therefore,in this paper,the RSSI(Received Signal Strength Indication)three-dimensional fingerprint orientation technology are analyzed and put forward improvement,respectively based on Wk NN(Weighted k-Nearest Neighbor)the fingerprint localization and fingerprint locating algorithm based on random forest is improved,and finally introduced ideas of fuzzy rules,the two improved algorithm of weighted positioning results together.Simulation experiments on UJIIndoor Loc,a publicly available Wi Fi fingerprint data set,show the feasibility and stability of the proposed algorithm.The most inquire about substance of this paper are as takes after:(1)In the process of RSSI fingerprint positioning,there are a variety of noise problems affecting the positioning results in the indoor environment.Using kalman filtering noise processing method is put forward,it is not directly to filter the noise of the RSSI,but only indirectly reduce the caused by the noise of the RSSI location estimation of volatility,and kalman filter in the case of unknown or time-varying noise,still can be used to filter the RSSI measurement noise filter,so choosing the kalman filter for noise processing improves the accuracy of the localization algorithm.(2)To balance the balance between computational complexity and estimation accuracy.Hierarchical clustering was selected and improved.An improved algorithm based on adaptive hierarchical clustering was proposed to classify fingerprint space,avoiding parameter setting and selection of initial clustering center.It can reduce the calculation and storage and improve the estimation accuracy.And because the relevant predefined subsets are not selected for the entire region,but are specific to each cluster,the improved adaptive hierarchical clustering allows the localization system to be extended to a larger region.(3)The fingerprint location algorithm based on Wk NN is analyzed and improved.Through the analysis of the fingerprint positioning algorithms that are widely used at present,the Wk NN algorithm with the highest positioning accuracy is selected and improved.For case,the k NN calculation specifically employments the normal arranges of different closest calibration focuses as hub position estimation,whereas the Wk NN calculation invests distinctive weights to diverse closest calibration focuses.Therefore,the node position estimated by Wk NN algorithm is usually more accurate than that estimated by k NN algorithm.On this basis,the enhanced Wk NN algorithm,which generates k value dynamically according to the threshold value,is proposed to improve the positioning effect.(4)The fingerprint location algorithm based on Random Forest is analyzed and improved.Many existing Wi Fi fingerprint indoor positioning systems need to collect a large amount of data in a specific environment when establishing a fingerprint database.Therefore,choose suitable for processing large amounts of data and good robustness of random forest algorithm,an improved algorithm of fingerprint orientation based on random forest,respectively by using the random forest classification and regression model was carried out on the floor level and the level of latitude and longitude data set training and learning from the floor and eventually level of latitude and longitude position respectively,to save storage space and have good anti-multipath characteristic.At the same time,the positioning performance of the algorithm under different system parameters,algorithms and input data sets is studied.(5)In order to achieve a better solution of location estimation,a rangeless location method based on Wk NN location and random forest location was proposed by introducing the idea of fuzzy rules.The fuzzy rule system was used to calculate the edge weight of the test node,and the estimated position was calculated according to the weighted average formula.Compared with traditional methods,fuzzy logic has many advantages,such as robustness,which enables it to deal with uncertainty in the environment and makes it possible to infer the user's location without a large number of samples.At long last,the execution of the progressed calculation is assessed and compared with a few comparable strategies,which proves the feasibility and superiority of the improved algorithm.